Research map / Multi-agent systems and games / Games and equilibria
Learning Global Nash Equilibrium in Team Competitive Games with Generalized Fictitious Cross-Play
Zelai Xu, Chao Yu, Yancheng Liang, Yi Wu, Yu Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add GFXP counter population instead of SP | MAgent Battle 3-vs-3 agents · MAgent Battle 3-vs-3 | exploitability: 28.66 -> 10.62 | Improved |
| add GFXP counter population instead of SP | MAgent Battle 10-vs-5 agents · MAgent Battle 10-vs-5 | exploitability: 42.21 -> 18.92 | Improved |
| use GFXP instead of PSRONash | MAgent Battle agents · MAgent Battle 10-vs-5 | exploitability: 44.88 -> 18.92 | Improved |
| use GFXP instead of ODO | MAgent Battle agents · MAgent Battle 3-vs-3 | exploitability: 21.76 -> 10.62 | Improved |
| use GFXP instead of SP | GRF Academy agents · GRF Academy PS 2-vs-1 | exploitability: 0.75 -> 0.35 | Improved |
| use GFXP instead of ODO (strongest baseline) | GRF Academy agents · GRF Academy HCA 4-vs-2 | exploitability: 0.44 -> 0.33 | Improved |
| use GFXP instead of ODO | GRF Academy agents · GRF Academy RPS 2-vs-1 | exploitability: 0.71 -> 0.60 | Improved |
| add reset (retrain from scratch each iteration) to GFXP (ablation) | GFXP · SAD game | steps to converge: 523 -> 943 | Hurt |
| remove warm start (reset) in PSRO and ODO (ablation) | PSRO and ODO · SAD game | convergence to global NE: converge to global NE -> degenerate to SP-like performance | Hurt |
| larger self-play ratio eta (0.8, 1) (ablation) | GFXP · SAD game | convergence to global NE: eta=0.8, 1 converge to local NE; eta=0, 0.2, 0.4, 0.6 converge to global NE | Hurt |
| use decaying eta (linear 1 to 0 in 5 iterations) instead of fixed eta (ablation) | GFXP · SAD game | steps to converge | Improved |
| increase game scale (actions A = 5, 10, 15, 20) (ablation) | GFXP · SAD game | steps to converge: A = 5, 10, 15, 20; converges in all runs | Mixed |
Closest papers by meaning
- Offline Fictitious Self-Play for Competitive Games (AAAI 2026 · similarity 0.91)
- Team-Fictitious Play for Reaching Team-Nash Equilibrium in Multi-team Games (NeurIPS 2024 · similarity 0.90)
- Accelerate Multi-Agent Reinforcement Learning in Zero-Sum Games with Subgame Curriculum Learning (AAAI 2024 · similarity 0.89)
- Learning Equilibria in Adversarial Team Markov Games: A Nonconvex-Hidden-Concave Min-Max Optimization Problem (NeurIPS 2024 · similarity 0.88)
- Provable Memory Efficient Self-Play Algorithm for Model-free Reinforcement Learning (ICLR 2024 · similarity 0.88)
- Mastering Zero-Shot Interactions in Cooperative and Competitive Simultaneous Games (ICML 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
